Abstract: In response to the challenges of on-orbit fault diagnosis for reaction flywheels in spacecraft attitude control systems, this paper conducts analysis based on in-flight data and proposes an on-orbit real-time diagnosis method for flywheels centered on a rapid estimation algorithm for equivalent friction torque. The core concept of this method is to uniformly model various faults that are difficult to directly measure in flywheels as anomalies in equivalent friction torque. An extended state observer is constructed to achieve rapid online estimation of such anomalies. Subsequently, the estimated equivalent friction torque is integrated with key directly measured flywheel variables—such as rotational speed, motor current, and bearing temperature—and input into a radial basis function neural network for fault classification. This enables the precise identification of specific fault types. Simulation results demonstrate that the proposed method offers good diagnostic accuracy and real-time performance, making it suitable for on-orbit flywheel condition monitoring and fault warning. It holds significant potential for engineering applications.